OpenAI Acquires Multi

> What if desktop computers were inherently multiplayer? What if the operating system placed people on equal footing to apps? Those were the questions we explored in building Multi, and before that, Remotion.

I presume that Multi actually did something useful, perhaps some sort of virtualization. But this description doesn't tell me anything useful about the company, nor does it even make sense. Is a desktop computer a video game? Why would it "play" let alone be multiplayer? Why would the OS be on equal footing to the apps? It doesn't make sense let alone be useful.

OpenAI stands to introduce the functionality of Multi into their own product. OpenAI is going to want control of your computer. It will want access to the data on your computer, to your screens, to what you're doing from moment to moment, and it wants to control it all for you. That's what it means by multiplayer.

If you grant it this control, you stand to lose any shred of privacy you have left, and become a complete slave to them. This is different from using assistants with granular access to data. It also is different from running your own private AI.

I wish Alexander and the rest of the Multi team all the best at OpenAI. I'm saddened by the loss of what I consider the best remote pair programming app around (if Mac-only is acceptable). Other than Tuple.app, are there any strong competitors? I tried Datadog's CoScreen early on, and the amount of chrome it added to windows made things a little janky (perhaps it's better now).

Anyone else think 30 days is just not enough notice before shutting off access for a paid app?

I totally understand free users getting a shorter notice period, but there's almost no instance where a paid user should get <60 days notice before being forced to migrate off a product they're using.

This sort of thing leads to a lot of mistrust in trying / using a startup for anything remotely important.

It became clear to me in the GPT-4o launch that OAI was interested in the "GitHub copilot for everything" route. With low-latency voice mode and the ability to take action on a user's behalf, this will basically feel like pair-programming for everything, or just having an employee who can do everything for you until they need your help or clarification.

To power that experience, an app will need to feel "multiplayer" like someone else is working with you. They'll probably bundle this in the API and have "agent mode" that developers can embed in any app or website, or just let consumers give OAI access to control their desktop. It'll also likely work async, so you can assign tasks, walk away or go to sleep for a few hours, and see the results.

This is speculation. But it feels like the interface that we'll look back on in ten years and say "that seemed obvious in hindsight."

I have been working on a similar space for the last 8 months. Creating a way for the AI use your OS. There are many projects out there trying to achieve this: Open Interpreter, OpenAdapt, etc.

The thing that is hard to achieve is not to get your LLM to operate your computer but to make it do the right thing. And by right thing I mean, the thing you expect it to do in the way you want. Example I always show to people when I give a demo of what I built: Say you want to buy a new Macbook. If you ask the computer to buy you one, it might:

A) Go to Google and search for a Macbook B) Go to Apple and search for a Macbook C) Go to Amazon and search for a Macbook

Now, depending on your implementation it might then ask you to give you more details about the Macbook you need (pro? air? how big?)

The tough part is to wrap all this together in a way that doesn't frustrate the user. In a way it's like when you do a Google search: the first result is the best guess of what you want. But how likely is it that it's not? Pretty much 50% of the time you look at the top 2-3 results right? Well it's the same with having the AI control your computer, at this point in time, you want to have "multiple sessions" - sort of like a parallel universe where the AI has done 2-3 things and lets you pick from A, B ... or C and then go from there.

Another approach would be to collate a library of most used "prompts" - i.e. if you are trying to buy a laptop and someone else managed to get the AI to go through a very thorough process where you're asked all the important details that lead to purchasing the Macbook, then you should be able to re-use that one workflow from the library, rather than have the AI start from scratch.

Anyway, it's very very tough - more so from a UX experience rather than an AI perspective.

"OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity."

Is this like EA where they have to make a trillion dollars first before getting to the actual mission?

I wonder what the play is here. Any ideas?

The key must be somewhere in the statement "we’ve been increasingly asking ourselves how we should work with computers. Not on or using computers, but truly with computers."

I wish I had some experience using Multi so I could picture this better. But is there a chance this is for a sandboxed execution environment that would allow models to interact with software alongside a human counterpart?

Is this acquisition or acqui-hire? Since they are shutting down their product sounds like it is former.

Sounds like investors are asking "What is the product you have?" and military ties and mixed people/AI chat apps are the answer.

Still waiting for the "PhD level" Q*.

Wow, shutting down a startup with a deadline being in less than a month that it was acquired without warning, that is a whole new reason to not trust startups at all.

Totally unprofessional.

I'm rather surprised that Keybase is still running after the team joined Zoom.

What an incredible journey.

Has OpenAI already reached that point in the lifecycle where they start gobbling up smaller companies? Damn this cycle is starting to feel faster and faster.

I have criticisms of OpenAI, but on the whole I really hope this doesn't mean they are losing the war and turning to diversification and other classic "big tech" moves. GPT still is (IME) the best at not assuming/inferring things in the prompt that aren't there (especially then mangling the prompt *cough* Gemini *cough*). For those of us who try to be very precise with our prompts, that is a big deal.

> We’ve closed new team signups, and currently active teams will be able to use the app until July 24, 2024, after which we’ll delete all user data.

What is it with OpenAI acquisitions reaming their user bases on short notice?

Is OpenAI already locked in such a dominant position that they don't have to pretend to care what anyone in industry thinks about them?

Everyone savvy knows that Microsoft and Oracle don't care, and are just waiting until they can stab you again. Even Google, who routinely pulls the rug out from under its customers, is gentler about it. Even the usual serial-founder brogrammer startups often have a semi-reasonable migration.

Sustaining engineering for legacy customers doesn't have to be expensive, and you can think of it as an obligation to pay back for your success, and maybe a foot in the door for sales.

Incredible how fast successful startups shut the door in front of their customers.

> Unfortunately, this means we’re sunsetting Multi. We’ve closed new team signups, and currently active teams will be able to use the app until July 24, 2024, after which we’ll delete all user data.

Seems a little abrupt.

OpenAI transformation from research lab to typical product company is almost complete.

Rockset acquisition announced 2 days ago. Now Multi.

OpenAI on an M&A spree.

Interesting.